Engineering teams have been running retrospectives for years, yet the way they run them keeps shifting. Smaller groups skip the long ceremony, remote teams write more and talk less, and platform and machine learning teams bring their own concerns to the table. This piece takes stock of where the start-stop-continue format stands today, what teams are doing with it, and which changes are worth copying. It is opinionated, short on theory and built around practices you can adopt this month.
Trend 1: shorter sessions, sharper lists
The all-afternoon retro is fading. Teams that once blocked half a day now run thirty to forty-five minutes with a firm cut-off. The reasoning is practical: a long session produces long lists, and long lists produce no action. Start-stop-continue helps here because its three columns impose a natural limit. Nobody can fit a sprawling post-mortem into three short lists, so the format forces prioritisation.
If your sessions still overrun, cap the lists at three entries each and make the facilitator enforce it. You will lose some detail and gain a lot of follow-through.
Trend 2: asynchronous input before the meeting
Distributed teams have made written pre-work normal. Each person adds entries to a shared page over a day or two, and the live session is spent grouping, discussing and choosing. This helps quieter members, avoids anchoring on the first speaker, and leaves a durable record. The practical tip is to open the page right after the work finishes, while memories are fresh, and close it twenty-four hours before the meeting.
Trend 3: entries tied to real events
Vague entries are out of fashion. Teams now attach a concrete moment to each item: the deploy that needed three attempts, the alert that fired at the wrong hour, the model release that waited on a data-labelling handoff. That discipline matters because a moment can be investigated, and a mood cannot. A retro entry should be something a teammate could verify by looking at a ticket or a log.
Trend 4: machine learning and operations teams adopt the format
Teams working on model delivery and incident handling have found the three-column structure fits their rhythm. A group that ships models might Start versioning evaluation datasets alongside code, Stop promoting models on the strength of a single test run, and Continue holding a short pre-release review. For background on how these teams plan their work, this planning guide aimed at data scientists and the engineers who ship their models gives useful context for what their lists often concern.
Operations teams are using it after incidents too. An on-call group might Start writing the first-hour timeline as events unfold, Stop paging three people for one alert, and Continue the habit of announcing hand-offs in the shared channel. For a perspective on how teams think about faster recovery, see this write-up on how operations teams approach quicker fixes for technical faults.
Trend 5: owners and dates by default
The most noticeable improvement in recent practice is that entries leave the room with owners attached. Teams have learned, often painfully, that a list without owners decays within two weeks. The common pattern now is to assign one person and one date to each entry before closing the meeting, then check the list at the start of the next session. It is a small procedural change with a large effect on completion.
Trend 6: looking outward for structure
Some teams deliberately borrow ideas from unrelated fields to freshen their format. One example some engineers mention is browsing websites that organise a big subject into approachable sections, such as BestPilotsNow, which presents itself as an aviation discovery platform. The takeaway is about presentation, not content: when information is cut into clearly labelled areas, people find their starting point faster, and the same applies to a retro page with three obvious columns.
Trend 7: separating sprint retros from incident reviews
Teams used to cram everything into one meeting. Increasingly they keep two rituals: a regular retro on how the work flowed, and a distinct review after any serious outage. The reasoning is that incidents deserve a careful timeline and a calm, blameless tone, while a retro is quicker and more general. Start-stop-continue belongs naturally in the regular retro, though some teams borrow its three columns at the end of an incident review to turn findings into changes.
The practical advice is to avoid mixing the two. A sprint retro that becomes an outage autopsy loses the ordinary friction that, left alone, causes the next outage.
What is not working
A few habits persist even though they rarely help.
- Retros without follow-up. If last time’s entries are never reviewed, the session turns into theatre.
- Tool-first rituals. Buying a retro platform before the team has a habit usually produces unused subscriptions. A shared document is enough.
- Blame in disguise. Entries that name individuals, however politely, chill the room. Keep them about practices.
- Endless parking lots. A backlog of “later” items nobody revisits becomes a graveyard. Prune it every quarter.
Examples of entries that earned their place
Concrete samples show the difference between a useful line and a vague one. Useful: “Start tagging each deployment with the ticket number so rollbacks are traceable.” Useful: “Stop reviewing pull requests larger than a few hundred lines in a single sitting.” Useful: “Continue the Thursday dependency-update slot, which kept upgrades small.” Vague: “Be more careful with releases.” The difference is that a reader can picture the first three being done on a given day, while the fourth gives no clue where to begin.
Who should run the session
A rotating facilitator is now the norm in healthy teams. The role carries few privileges, mainly keeping time, protecting quiet voices and recording owners, but rotating it spreads the habit of listening. A manager can attend without leading, which reduces the pressure on honest entries. Some teams invite an engineer from a neighbouring group as a guest every few months, because an outsider asks the naive questions that expose assumptions nobody inside the team still notices.
Whoever runs it, make the expectations clear at the start: the page is about practices, the lists are short, and every entry gets an owner or gets dropped. Repeating these three rules each time takes ten seconds and prevents most of the drift described above.
A practical template for this month
If you want to modernise your own sessions without a big programme, try this arrangement for the next two sprints.
- Open a shared page the day the sprint ends and invite written entries for twenty-four hours.
- Run a forty-minute session: ten minutes reading, twenty minutes grouping and choosing, ten minutes assigning owners.
- Limit each column to three entries, each tied to a specific event.
- Put the page in the team channel and review it for two minutes at the next sprint’s start.
That is the whole arrangement. It needs no software purchase, no facilitator training and no approval from anyone outside the team.
Where this is heading
Looking ahead, a pair of shifts deserve attention. First, teams are increasingly linking retro entries to the tickets that implement them, so the connection between a lesson and the work it triggered is visible. Second, more groups are comparing their lists over time, noticing that a Start entry which recurs three times signals a problem too large to tackle in one step.
Neither requires new tooling. Both come down to treating the retrospective as a running record rather than a series of unrelated meetings. Groups that work this way usually see their recurring friction points shrink, and that the Continue column becomes a quiet statement of what the team has agreed is good practice.